Two-stage pedestrian searching method combining face and appearance

Abstract

The invention provides a two-stage pedestrian searching method combining face and appearance. The method comprises: detecting faces of pedestrians by utilizing a face detector, and outputting a face representation vector based on a face comparison model; sorting the to-be-matched pedestrian set according to the Euclidean distance to obtain a face sorting result; selecting a plurality of samples asmulti-matching targets of the next stage according to a sorting result; taking the selected next-stage multi-matching targets of different pedestrians in the same data set as negative samples of eachother, so as to compress the next-stage matching sample space; and finally, carrying out multi-target pedestrian re-identification, sorting the to-be-matched image set according to the average distance or similarity with the plurality of targets, and outputting a final result. According to the method, the face and pedestrian re-recognition are combined through the convolutional neural network based on the DenseNet to search the target pedestrian, the robustness is improved through the elaborately designed limiting conditions, the calculation amount is reduced, and the pedestrian searching performance is further improved.

Description

technical field [0001] The invention relates to the technical field of computer vision, in particular to a two-stage pedestrian search method combining human face and appearance. Background technique [0002] At present, person re-identification (Person Re-identification) technology is playing an increasingly important role in practical applications, such as security monitoring, crowd flow monitoring, and pedestrian behavior analysis. Nowadays, most computer vision researchers focus on pedestrian re-identification research on pedestrian matching between multiple cameras in similar scenes. [0003] However, in practical applications, for example, when monitoring criminal suspects, there is often a large difference in appearance between the target pedestrian image (such as a portrait photo) and the pedestrian image captured by the camera to be matched, and the traditional pedestrian re-identification method cannot be effective. accurately match the target pedestrian. The mor...

AI Technical Summary

Problems solved by technology

[0003] However, in practical applications, for example, when monitoring criminal suspects, there is often a large difference in appearance between the target pedestrian image (such as a portrait photo) and the pedestrian image captured by the camera to be matched, and the traditional pedestrian re-identification method cannot be effective. match the target pedestrian

The more reliable pedestrian feature in this scene should be the face feature, but if only the face is used for matching, the target pedestrian sample without a face cannot be matched, and the tracking of its trajectory may be lost

Images